{
  "project": "PhysLayer",
  "paper": "https://arxiv.org/abs/2604.23574",
  "source": "Author-supplied ICME 2026 manuscript. Values are transcribed, not independently reproduced.",
  "missing_values": null,
  "tables": {
    "main": {
      "source": "Main paper, Table 1",
      "metrics": [
        "CLIP similarity",
        "FID",
        "Motion-FID",
        "Motion smoothness"
      ],
      "directions": [
        "higher",
        "lower",
        "lower",
        "higher"
      ],
      "precision": [
        2,
        2,
        2,
        3
      ],
      "rows": [
        [
          "PIA",
          16.17,
          215.06,
          418.51,
          0.981
        ],
        [
          "DynamiCrafter",
          16.16,
          185.86,
          429.22,
          0.954
        ],
        [
          "CogVideoX",
          16.18,
          120.91,
          406.32,
          0.978
        ],
        [
          "I2VGen-XL",
          16.28,
          187.35,
          442.91,
          0.956
        ],
        [
          "PhysGen",
          null,
          112.83,
          401.61,
          0.982
        ],
        [
          "PhysLayer",
          16.64,
          102.32,
          389.64,
          0.986
        ]
      ]
    },
    "ablation": {
      "source": "Main paper, Table 2",
      "metrics": [
        "CLIP similarity",
        "FID",
        "Motion-FID",
        "Motion smoothness"
      ],
      "directions": [
        "higher",
        "lower",
        "lower",
        "higher"
      ],
      "precision": [
        2,
        2,
        2,
        3
      ],
      "rows": [
        [
          "Without depth scaling",
          16.52,
          118.45,
          405.28,
          0.979
        ],
        [
          "Without depth motion",
          16.48,
          115.67,
          398.42,
          0.981
        ],
        [
          "Without relighting",
          16.61,
          108.94,
          392.11,
          0.985
        ],
        [
          "Without language guidance",
          15.93,
          105.28,
          391.87,
          0.984
        ],
        [
          "Full PhysLayer",
          16.64,
          102.32,
          389.64,
          0.986
        ]
      ]
    },
    "human": {
      "source": "Main paper, Table 3",
      "metrics": [
        "Text\u2013video alignment",
        "Visual quality",
        "Physical plausibility"
      ],
      "directions": [
        "higher",
        "higher",
        "higher"
      ],
      "precision": [
        2,
        2,
        2
      ],
      "rows": [
        [
          "PIA",
          1.86,
          1.68,
          1.55
        ],
        [
          "DynamiCrafter",
          1.97,
          2.02,
          1.82
        ],
        [
          "I2VGen-XL",
          2.21,
          2.54,
          2.16
        ],
        [
          "CogVideoX",
          3.12,
          3.35,
          2.63
        ],
        [
          "PhysGen",
          null,
          3.56,
          3.31
        ],
        [
          "PhysLayer",
          4.23,
          3.83,
          4.1
        ]
      ]
    }
  }
}